hubert_arabic_mdd_5

This model is a fine-tuned version of facebook/hubert-large-ls960-ft on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6845
• Wer: 0.1910
• Per: 0.1682

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Per
21.0868	1.0	82	3.5829	1.0	1.0
3.5419	2.0	164	3.3544	1.0	1.0
3.3109	3.0	246	3.1878	1.0	1.0
3.2117	4.0	328	3.1324	1.0	1.0
3.0509	5.0	410	2.9238	1.0	1.0
2.7904	6.0	492	2.4830	1.0	1.0
2.2728	7.0	574	2.0478	0.8358	0.8474
1.8365	8.0	656	1.6043	0.7094	0.7206
1.4603	9.0	738	1.2872	0.6193	0.6247
1.2203	10.0	820	1.2694	0.4343	0.4229
0.9726	11.0	902	0.9350	0.4191	0.4116
0.8458	12.0	984	0.8471	0.3734	0.3640
0.7365	13.0	1066	0.8078	0.3530	0.3439
0.6204	14.0	1148	0.7618	0.3031	0.2874
0.5913	15.0	1230	0.7372	0.2598	0.2382
0.5217	16.0	1312	0.7219	0.2354	0.2149
0.4866	17.0	1394	0.7125	0.2071	0.1836
0.4735	18.0	1476	0.7071	0.1967	0.1742
0.462	19.0	1558	0.6910	0.1910	0.1677
0.458	20.0	1640	0.6845	0.1910	0.1682

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0